osdir.com

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Process JobBundleFactory for portable runner


I would see support for staging libraries as optional / nice to have since that can also be done as part of host provisioning (i.e. in the Python case a virtual environment was already setup and just needs to be activated).

Depending on how the command that launches the harness is configured, additional steps such as virtualenv activate or setting of other environment variables can be included as well.


On Thu, Aug 23, 2018 at 5:15 AM Maximilian Michels <mxm@xxxxxxxxxx> wrote:
Just to recap:

 From this and the other thread ("Bootstraping Beam's Job Server") we
got sufficient evidence that process-based execution is a desired feature.

Process-based execution as an alternative to dockerized execution
https://issues.apache.org/jira/browse/BEAM-5187

Which parts are executed as a process?
=> The SDK harness for user code

What configuration options are supported?
=> Provide information about the target architecture (OS/CPU)
=> Staging libraries, as also supported by Docker
=> Activating a pre-existing environment (e.g. virutalenv)


On 23.08.18 14:13, Maximilian Michels wrote:
>> One thing to consider that we've talked about in the past. It might
>> make sense to extend the environment proto and have the SDK be
>> explicit about which kinds of environment it support
>
> +1 Encoding environment information there is a good idea.
>
>> Seems it will create a default docker url even if the
>> hardness_docker_image is set to None in pipeline options. Shall we add
>> another option or honor the None in this option to support the process
>> job?
>
> Yes, if no Docker image is set the default one will be used. Currently
> Docker is the only way to execute pipelines with the PortableRunner. If
> the docker_image is not set, execution won't succeed.
>
> On 22.08.18 22:59, Xinyu Liu wrote:
>> We are also interested in this Process JobBundleFactory as we are
>> planning to fork a process to run python sdk in Samza runner, instead
>> of using docker container. So this change will be helpful to us too.
>> On the same note, we are trying out portable_runner.py to submit a
>> python job. Seems it will create a default docker url even if the
>> hardness_docker_image is set to None in pipeline options. Shall we add
>> another option or honor the None in this option to support the process
>> job? I made some local changes right now to walk around this.
>>
>> Thanks,
>> Xinyu
>>
>> On Tue, Aug 21, 2018 at 12:25 PM, Henning Rohde <herohde@xxxxxxxxxx
>> <mailto:herohde@xxxxxxxxxx>> wrote:
>>
>>     By "enum" in quotes, I meant the usual open URN style pattern not an
>>     actual enum. Sorry if that wasn't clear.
>>
>>     On Tue, Aug 21, 2018 at 11:51 AM Lukasz Cwik <lcwik@xxxxxxxxxx
>>     <mailto:lcwik@xxxxxxxxxx>> wrote:
>>
>>         I would model the environment to be more free form then enums
>>         such that we have forward looking extensibility and would
>>         suggest to follow the same pattern we use on PTransforms (using
>>         an URN and a URN specific payload). Note that in this case we
>>         may want to support a list of supported environments (e.g. java,
>>         docker, python, ...).
>>
>>         On Tue, Aug 21, 2018 at 10:37 AM Henning Rohde
>>         <herohde@xxxxxxxxxx <mailto:herohde@xxxxxxxxxx>> wrote:
>>
>>             One thing to consider that we've talked about in the past.
>>             It might make sense to extend the environment proto and have
>>             the SDK be explicit about which kinds of environment it
>>             supports:
>>
>>             
>> https://github.com/apache/beam/blob/8c4f4babc0b0d55e7bddefa3f9f9ba65d21ef139/model/pipeline/src/main/proto/beam_runner_api.proto#L969
>>
>>             
>> <https://github.com/apache/beam/blob/8c4f4babc0b0d55e7bddefa3f9f9ba65d21ef139/model/pipeline/src/main/proto/beam_runner_api.proto#L969>
>>
>>
>>             This choice might impact what files are staged or what not.
>>             Some SDKs, such as Go, provide a compiled binary and _need_
>>             to know what the target architecture is. Running on a mac
>>             with and without docker, say, requires a different worker in
>>             each case. If we add an "enum", we can also easily add the
>>             external idea where the SDK/user starts the SDK harnesses
>>             instead of the runner. Each runner may not support all types
>>             of environments.
>>
>>             Henning
>>
>>             On Tue, Aug 21, 2018 at 2:52 AM Maximilian Michels
>>             <mxm@xxxxxxxxxx <mailto:mxm@xxxxxxxxxx>> wrote:
>>
>>                 For reference, here is corresponding JIRA issue for this
>>                 thread:
>>                 https://issues.apache.org/jira/browse/BEAM-5187
>>                 <https://issues.apache.org/jira/browse/BEAM-5187>
>>
>>                 On 16.08.18 11:15, Maximilian Michels wrote:
>>                  > Makes sense to have an option to run the SDK harness
>>                 in a non-dockerized
>>                  > environment.
>>                  >
>>                  > I'm in the process of creating a Docker entry point
>>                 for Flink's
>>                  > JobServer[1]. I suppose you would also prefer to
>>                 execute that one
>>                  > standalone. We should make sure this is also an
>> option.
>>                  >
>>                  > [1] https://issues.apache.org/jira/browse/BEAM-4130
>>                 <https://issues.apache.org/jira/browse/BEAM-4130>
>>                  >
>>                  > On 16.08.18 07:42, Thomas Weise wrote:
>>                  >> Yes, that's the proposal. Everything that would
>>                 otherwise be packaged
>>                  >> into the Docker container would need to be
>>                 pre-installed in the host
>>                  >> environment. In the case of Python SDK, that could
>>                 simply mean a
>>                  >> (frozen) virtual environment that was setup when the
>>                 host was
>>                  >> provisioned - the SDK harness process(es) will then
>>                 just utilize that.
>>                  >> Of course this flavor of SDK harness execution could
>>                 also be useful in
>>                  >> the local development environment, where right now
>>                 someone who already
>>                  >> has the Python environment needs to also install
>>                 Docker and update a
>>                  >> container to launch a Python SDK pipeline on the
>>                 Flink runner.
>>                  >>
>>                  >> On Wed, Aug 15, 2018 at 12:40 PM Daniel Oliveira
>>                 <danoliveira@xxxxxxxxxx <mailto:danoliveira@xxxxxxxxxx>
>>                  >> <mailto:danoliveira@xxxxxxxxxx
>>                 <mailto:danoliveira@xxxxxxxxxx>>> wrote:
>>                  >>
>>                  >>      I just want to clarify that I understand this
>>                 correctly since I'm
>>                  >>      not that familiar with the details behind all
>>                 these execution
>>                  >>      environments yet. Is the proposal to create a
>>                 new JobBundleFactory
>>                  >>      that instead of using Docker to create the
>>                 environment that the new
>>                  >>      processes will execute in, this
>>                 JobBundleFactory would execute the
>>                  >>      new processes directly in the host environment?
>>                 So in practice if I
>>                  >>      ran a pipeline with this JobBundleFactory the
>>                 SDK Harness and Runner
>>                  >>      Harness would both be executing directly on my
>>                 machine and would
>>                  >>      depend on me having the dependencies already
>>                 present on my machine?
>>                  >>
>>                  >>      On Mon, Aug 13, 2018 at 5:50 PM Ankur Goenka
>>                 <goenka@xxxxxxxxxx <mailto:goenka@xxxxxxxxxx>
>>                  >>      <mailto:goenka@xxxxxxxxxx
>>                 <mailto:goenka@xxxxxxxxxx>>> wrote:
>>                  >>
>>                  >>          Thanks for starting the discussion. I will
>>                 be happy to help.
>>                  >>          I agree, we should have pluggable
>>                 SDKHarness environment Factory.
>>                  >>          We can register multiple Environment
>>                 factory using service
>>                  >>          registry and use the PipelineOption to pick
>>                 the right one on per
>>                  >>          job basis.
>>                  >>
>>                  >>          There are a couple of things which are
>>                 require to setup before
>>                  >>          launching the process.
>>                  >>
>>                  >>            * Setting up the environment as done in
>>                 boot.go [4]
>>                  >>            * Retrieving and putting the artifacts in
>>                 the right location.
>>                  >>
>>                  >>          You can probably leverage boot.go code to
>>                 setup the environment.
>>                  >>
>>                  >>          Also, it will be useful to enumerate pros
>>                 and cons of different
>>                  >>          Environments to help users choose the right
>>                 one.
>>                  >>
>>                  >>
>>                  >>          On Mon, Aug 6, 2018 at 4:50 PM Thomas Weise
>>                 <thw@xxxxxxxxxx <mailto:thw@xxxxxxxxxx>
>>                  >>          <mailto:thw@xxxxxxxxxx
>>                 <mailto:thw@xxxxxxxxxx>>> wrote:
>>                  >>
>>                  >>              Hi,
>>                  >>
>>                  >>              Currently the portable Flink runner
>>                 only works with SDK
>>                  >>              Docker containers for execution
>>                 (DockerJobBundleFactory,
>>                  >>              besides an in-process (embedded)
>>                 factory option for testing
>>                  >>              [1]). I'm considering adding another
>>                 out of process
>>                  >>              JobBundleFactory implementation that
>>                 directly forks the
>>                  >>              processes on the task manager host,
>>                 eliminating the need for
>>                  >>              Docker. This would work reasonably well
>>                 in environments
>>                  >>              where the dependencies (in this case
>>                 Python) can easily be
>>                  >>              tied into the host deployment (also
>>                 within an application
>>                  >>              specific Kubernetes pod).
>>                  >>
>>                  >>              There was already some discussion about
>>                 alternative
>>                  >>              JobBundleFactory implementation in [2].
>>                 There is also a JIRA
>>                  >>              to make the bundle factory pluggable
>>                 [3], pending
>>                  >>              availability of runner level options.
>>                  >>
>>                  >>              For a "ProcessBundleFactory", in
>>                 addition to the Python
>>                  >>              dependencies the environment would also
>>                 need to have the Go
>>                  >>              boot executable [4] (or a substitute
>>                 thereof) to perform the
>>                  >>              harness initialization.
>>                  >>
>>                  >>              Is anyone else interested in this SDK
>>                 execution option or
>>                  >>              has already investigated an alternative
>>                 implementation?
>>                  >>
>>                  >>              Thanks,
>>                  >>              Thomas
>>                  >>
>>                  >>              [1]
>>                  >>
>>                 
>> https://github.com/apache/beam/blob/7958a379b0a37a89edc3a6ae4b5bc82fda41fcd6/runners/flink/src/test/java/org/apache/beam/runners/flink/PortableExecutionTest.java#L83
>>
>>                 
>> <https://github.com/apache/beam/blob/7958a379b0a37a89edc3a6ae4b5bc82fda41fcd6/runners/flink/src/test/java/org/apache/beam/runners/flink/PortableExecutionTest.java#L83>
>>
>>                  >>
>>                  >>              [2]
>>                  >>
>>                 
>> https://lists.apache.org/thread.html/d6b6fde764796de31996db9bb5f9de3e7aaf0ab29b99d0adb52ac508@%3Cdev.beam.apache.org%3E
>>
>>                 
>> <https://lists.apache.org/thread.html/d6b6fde764796de31996db9bb5f9de3e7aaf0ab29b99d0adb52ac508@%3Cdev.beam.apache.org%3E>
>>
>>                  >>
>>                  >>              [3]
>>                 https://issues.apache.org/jira/browse/BEAM-4819
>>                 <https://issues.apache.org/jira/browse/BEAM-4819>
>>                  >>
>>                  >>              [4]
>>                 
>> https://github.com/apache/beam/blob/master/sdks/python/container/boot.go
>>                 
>> <https://github.com/apache/beam/blob/master/sdks/python/container/boot.go>
>>
>>                  >>
>>
>>                 --                 Max
>>
>>
>

--
Max